Discovery of Sustainable Refrigerants through Physics-Informed RL Fine-Tuning of Sequence Models

2025-09-23

AI Summary

Conventional refrigerants—particularly hydrofluorocarbons (HFCs)—are potent greenhouse gases, necessitating environmentally benign alternatives. However, the known candidate molecular space remains extremely limited (~300 compounds), with few newly experimentally validated molecules and severely scarce thermodynamic data, hindering purely data-driven discovery approaches. Method: We propose Refgen, a knowledge-integrated generative framework that jointly embeds thermodynamic priors—including equation-of-state constraints and full-cycle thermodynamic simulation—into a reinforcement learning fine-tuning pipeline. It synergistically combines sequence-based molecular modeling, property prediction, and physics-guided generation. Contribution/Results: Refgen enables physically consistent *de novo* molecular design under data scarcity, transcending conventional refrigerant chemical space boundaries. It significantly improves thermodynamic feasibility and environmental sustainability (e.g., low global warming potential, zero ozone depletion potential) of generated candidates. The framework establishes a scalable, interpretable, and physics-aware generative paradigm for green refrigerant discovery.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Most refrigerants currently used in air-conditioning systems, such as hydrofluorocarbons, are potent greenhouse gases and are being phased down. Large-scale molecular screening has been applied to the search for alternatives, but in practice only about 300 refrigerants are known, and only a few additional candidates have been suggested without experimental validation. This scarcity of reliable data limits the effectiveness of purely data-driven methods. We present Refgen, a generative pipeline that integrates machine learning with physics-grounded inductive biases. Alongside fine-tuning for valid molecular generation, Refgen incorporates predictive models for critical properties, equations of state, thermochemical polynomials, and full vapor compression cycle simulations. These models enable reinforcement learning fine-tuning under thermodynamic constraints, enforcing consistency and guiding discovery toward molecules that balance efficiency, safety, and environmental impact. By embedding physics into the learning process, Refgen leverages scarce data effectively and enables de novo refrigerant discovery beyond the known set of compounds.

Problem

• Finding sustainable refrigerants to replace potent greenhouse gases
• Overcoming data scarcity limitations in purely data-driven discovery methods
• Balancing efficiency, safety and environmental impact in refrigerant design

Innovation

• Physics-informed reinforcement learning fine-tuning
• Integration of thermodynamic constraints and simulations
• Generative molecular design with inductive biases